Formation Of Multi-dimensional Public Opinion In Sina Weibo And Its Application In Monitoring High-risk Users

With the rapid development of Internet technology,the speed of information dissemination and update is further accelerated.Sina Weibo,as an important social platform,frequently generates public opinion topics.In general,a discussion topic of hot event breaks outin Weibo and form initial public opinion,after further disclosure of information and communication between the user and to derive from more than one child topics related to the initial public opinion,department of molecular topic further evolution to form new derivative public opinion,and initial multidimensional public opinion,public opinion interweave together to form become more huge destructive power of public opinion.Based on this,this paper takes the topic derivative phenomenon of Sina Weibo as the entry point to further study the formation process of multi-dimensional public opinion and monitor the users who participate in public opinion discussions.The specific contents are as follows:(1)Modeling the formation process of multi-dimensional public opinion in Sina Weibo.First introduced the initial public opinion variation mechanism to reveal the process of derivative sub topics and then spreads out Brown movement is used to determine the item parameter and dynamics simulation model based on complex network spread its evolution rule,then introduce the basic reproductive number identifying whether derived sub topics to form a new dimension of public opinion,to reveal the multi-dimensional form the whole process of public opinion.Secondly,through simulation experiments,the influence of information alienation degree,environmental forces,topic correlation coefficient,the amount of information contained in sub-topics and network topology structure on the formation of multi-dimensional public opinion is discussed.(2)Research on user monitoring based on user portrait and random forest algorithm.Firstly,the hot topics of multi-dimensional public opinion events are selected,and the user data and comment text that participate in public opinion discussions are extracted by crawler to construct the index system of network public opinion user portrait,and the data and comment text are quantified into comprehensive indicators such as user activity,user influence and user emotional tendency.According to the size index sorting users involved in the discussions are divided into three types of high-risk,moderate and low risk,and then set up high-risk users recognition model based on random forest algorithm,selecting the sample training and build a random forest classifier precision testing,and training the classifier can be used to predict whether the newly released public opinion information users in high-risk users.Comprehensive the above model building and select two case to validate its respectively,the results show that:(1)sub topics include the amount of information and environmental forces are key factors affecting the formation of multidimensional public opinion,which had a greater influence on the environment forces the number of their topic,and sub topics included information is sub topics can form public opinion of the key factors.(2)Most users in online public opinions are ordinary users,and only a few users are high-risk users,and the recognition accuracy of random forest classifier is as high as 98.98%,which indicates that public opinion controllers can use this model to monitor a small number of high-risk users,thus greatly reducing the cost of public opinion control.